---
title: "Daryani et al.'s year-naming test does not reach ChronoCulture's older-year asymmetry — it tested only naming the most recent surveyed year, which made no real difference, leaving that one ChronoCulture claim single-sourced"
type: "claim"
status: "seedling"
audit_status: "capture-verified"
source_url: "https://arxiv.org/pdf/2609.01902"
source_title: "Accurate in space, unreliable in time: how LLMs represent national cultural change"
source_author: "Yalda Daryani, Miranda Bogen, Madeleine I. G. Daepp"
source_venue: "arXiv:2609.01902 [cs.CY]; Dept. of Psychology, University of Southern California, and Center for Democracy and Technology"
source_date: "2026-09-01T00:00:00.000Z"
source_tier: 1
source_sha: "a6d68d2c05374440732800dca5e7388925f3dd07419f3b8cf4e799906b92b445"
source_quote: "Naming the country alone (C1) put models about as close to the recent snapshot as naming the country and year together (C2)"
source_quote_2: "Naming the recent year does not pull a model forward out of the lag documented above, and the explicit date neither refines its answer nor moves it toward the surveyed position for that year."
source_quote_3: "close to zero for every model, from −0.005 for Claude to +0.009 for GPT"
independent_roots: 1
provenance: "Promotion from 10-inbox/raw/2026-10-18-the-chronoculture-branch-now-rests-three-claim-notes.md, 2026-10-24"
origin: "batch"
derived_from: ["20261018-0213-the-chronoculture-branch-now"]
date_created: "2026-10-24T00:00:00.000Z"
writer_model: "claude-opus-4-8"
tags: ["llm-evaluation","cultural-awareness","world-values-survey","corroboration","prompting","arxiv"]
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-25 — source_quote verbatim (normalized) in the capture-time archive (sha256 a6d68d2c0537…) [reading-order], same-night model-free sweep"
verified_verbatim: "2026-10-26 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


The corroboration that Daryani, Bogen & Daepp provide for the [[observation-daryani-independently-corroborates-chronoculture-discharging-single-source-cap|ChronoCulture branch]] stops short of one of its three claim-notes. [[claim-chronoculture-naming-older-survey-year-helps-only-as-real-shift-grows-larger|ChronoCulture's year-naming finding]] is a conditional asymmetry across *two* target years decades apart: naming the year helps near-zero on average, but naming the *older* of the two years shows a growing benefit as the real shift between them grows large, while naming the newer year does not.

Daryani et al. tested a different, narrower question. They compared naming the country alone against naming the country together with its *most recent* surveyed year, and asked whether the explicit recent date pulled a model's answer closer to that recent snapshot. It did not: naming the country alone put models about as close to the recent snapshot as naming the country and year together, with a difference close to zero for every model, from −0.005 for Claude to +0.009 for GPT. In their words, naming the recent year neither refines a model's answer nor moves it toward the surveyed position for that year.

This is directionally consistent with ChronoCulture's finding that year-naming is weak on average. But because Daryani et al. never varied a decades-distant *older* year under a *large* real shift — the exact condition under which ChronoCulture's benefit appears — it can neither confirm nor disconfirm that specific conditional effect. That piece of the branch therefore remains single-sourced on arXiv:2610.20472 alone (`independent_roots: 1`), and the open verification is tracked at [[question-corroborate-chronoculture-older-year-naming-asymmetry]].

> [!note] Seek's commentary:
> The easy write-up says "the branch is corroborated" and lets this third claim ride along on the other two. It can't. Daryani asked whether telling a model *now* helps it answer about now, and found it doesn't — a cousin of ChronoCulture's question, not the same one. ChronoCulture's live result is about telling a model *then*, across a wide gap, and watching the anchor finally bite. Nobody else has run that experiment. The most mechanism-flavored of the three claims — the one that hints at how these models default to a present-skewed average — is also the one still standing on a single preprint. Two out of three is the honest count.
> — Seek